Long conversation key information extraction method for 5G new conversation environment

Abstract

The invention relates to a long dialogue information extraction method for a 5G new call. The method comprises the following steps: 1, voice signal collection and preprocessing: a system collects voice signals between a first user and a second user through a 5G call device; 2, automatic speech recognition: converting the preprocessed speech signals into text data by an automatic speech recognition technology; thirdly, key information is extracted in an iteration mode; and 4, analyzing an output result. According to the method, through a multi-level key information extraction and iterative optimization mechanism, the information extraction capability of a large language model (LLM) in a long dialogue scene is remarkably enhanced. According to the method, the problems of complexity of long text understanding and limitation of computing resources are effectively relieved.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and specifically to a method for extracting key information of long conversations in a new 5G call environment. Background Art

[0002] At present, the long conversation information extraction technology for the new 5G call environment mainly involves key technologies such as automatic speech recognition (ASR), text summarization, keyword extraction, and to-do extraction. Since ASR is only the initial step in the information processing process, this article does not discuss it in depth. Text summarization methods include statistical extractive summarization and generative summarization. The statistical extractive summarization method generates a summary by extracting key sentences or phrases from the text, while the generative summarization method uses a language model to generate new sentences to construct a summary. The extractive summary has certain guarantees in grammar and syntax, but it also faces certain problems, such as incorrect content selection, poor coherence, and poor flexibility. The generative summary allows new words or phrases to be included in the summary, and has high flexibility. The Large Language Model (LLM) generates summaries that are highly fluent and coherent due to its powerful semantic understanding ability and flexible generation method. Keywords are words that can express the central content of a document. Traditional keyword extraction methods include frequency-based methods, graph-based methods, and statistical methods, which have played a foundational role in the field of natural language processing (NLP). However, the emergence of LLM has ushered in a new era of keyword extraction, which can accurately capture the nuances of context, semantic relationships, and the complexity of specific fields. LLM has enhanced its ability to capture context and semantics, making it more suitable for extracting keywords in complex language fields. A to-do list is a list of tasks or activities that need to be completed at a certain point in the future. Before the rise of LLM, to-do extraction used rule-based methods and text classification methods to identify to-dos. Rule-based methods are suitable for simple and structured scenarios, with limited applicability but strong controllability. Text classification-based methods use pre-trained models such as BERT and Fastext, and their performance depends largely on high-quality training data and is sensitive to new data. LLM can understand the context and handle complex and flexible to-do extraction with the best effect.

[0003] When designing LLM, developers usually set an upper limit on the length of the input text to balance the model's capabilities and computing performance. Therefore, when processing long texts, it is often necessary to truncate or split the text to meet the input length constraint. With the continuous development of LLM, the maximum number of tokens supported is also gradually increasing, from the initial 4K and 8K to the current 16K, 32K, and even 128K. This progress provides more powerful support for long text processing, but it also raises a series of new challenges:

[0004] 1. Computing resource limitations

[0005] Although the current LLM can theoretically support longer token sequences, computing resources (especially GPU memory) are still the main bottleneck in practical applications. For example, even if the input text is within the range of 128K tokens, the system may not be able to process the entire sample at once due to the limited memory capacity. Therefore, GPU memory capacity becomes a key limiting factor in long text processing.

[0006] 2. Insufficient model understanding

[0007] With sufficient resources, many open source LLMs still have shortcomings in understanding and summarizing long texts. Inputting too many tokens at one time may make it difficult for the model to accurately capture key information, or even produce "hallucination" phenomena. Therefore, how to ensure that the model can fully understand each fragment and effectively extract key information during the long text processing process is an important research direction. Summary of the invention

[0008] The purpose of the present invention is to provide a method for extracting key information of long conversations in a new 5G call environment to solve the problems raised in the above background technology.

[0009] To achieve the above object, the present invention provides the following technical solution: a method for extracting key information of a long conversation in a 5G new call environment, the method comprising the following steps:

[0010] Step 1: Voice signal collection and preprocessing: The system collects the voice signal between the first user and the second user through the 5G call device;

[0011] Step 2: Automatic speech recognition: Automatic speech recognition technology converts the preprocessed speech signal into text data;

[0012] Step 3: Extracting key information, the key information extraction includes the following sub-steps:

[0013] Sub-step 31: splitting the long text according to a preset segmentation standard to obtain a plurality of short text segments after the segmentation;

[0014] Sub-step 32: Input the short text segment and the first prompt word Prompt 1 into the large language model LLM to generate key information: the extraction results generated by the large language model LLM include to-do items, keywords and text summaries;

[0015] Sub-step 33: Extract results by iteration: Take partial results of the first stage and concatenate them with the second prompt word Prompt2, input them into the large language model LLM again, and generate the key information extraction results of the second stage. 1,1 ,…,R 1,m After splicing, input LLM with Prompt 2 to generate the key information extraction result R of the second stage 2,1 By analogy, R 1,n ,…,R 1,i After splicing, input LLM with Prompt 2 to get the result R of the second stage 2,j .

[0016] R 1，1 , …, R 1，m →R 2，1 ;…;R 1，n , …, R 1，i →R 2，j

[0017] In each iteration, the output of the previous stage (R n,i ) will be used as the input for the next stage, and after being concatenated with the second prompt word Prompt2, it will be input into the large language model LLM for processing again; this process will continue to iterate until the total length of the input text and the second prompt word Prompt2 can be input into the large language model LLM at one time, generating the final key information extraction result R;

[0018] Step 4: Output result analysis.

[0019] Preferably, the length of the short text segment is limited by the difference between the input length and the length of the first prompt word Prompt 1.

[0020] Preferably, the concatenation length is also limited by the difference between the input token length set by the system and the token length of Prompt2.

[0021] Preferably, the large language model LLM includes an open source model and / or a closed source model.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] The present invention can enhance the ability to understand long conversations, effectively overcoming the problem of insufficient understanding faced by the large language model (LLM) when processing long conversation inputs, and improves the processing accuracy of complex long texts through a multi-level information extraction and optimization mechanism.

[0024] At the same time, it optimizes the utilization of computing resources and solves the challenge of significantly increasing GPU computing resource requirements in the LLM reasoning phase as the number of input tokens increases, thereby improving the resource utilization efficiency of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of the present invention;

[0026] Figure 2 A specific example of Prompt 1 provided by the present invention;

[0027] Figure 3 A specific example of Prompt2 provided by the present invention; DETAILED DESCRIPTION

[0028] First, the terms involved in the embodiments of this specification are explained.

[0029] Text Summarization: The goal of text summarization is to extract the main ideas from one or more text sources and create a descriptive text that is short, coherent, and consistent with the original text.

[0030] Keyword extraction: Keyword extraction aims to identify and extract important words or phrases from a text that best represent the theme or core content of the text.

[0031] To-do extraction task in the field of natural language processing (NLP) refers to the process of automatically identifying and extracting clear tasks, action items or to-do items from unstructured text.

[0032] A Large Language Model (LLM) is an artificial intelligence model designed to understand and generate human language. LLMs usually refer to language models containing tens of billions (or more) of parameters, which are trained on massive amounts of text data to gain a deep understanding of language.

[0033] The technical solution in one embodiment of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0034] See also Figure 1-3 :A method for extracting key information of long conversations in a 5G new call environment, the method comprising the following steps:

[0035] Step 1: Speech signal acquisition and preprocessing:

[0036] The system collects the voice signal between the first user and the second user through the 5G call device. During the signal collection stage, a series of pre-processing techniques are applied, including denoising, echo cancellation and gain control, to improve the accuracy of the subsequent automatic speech recognition (ASR) process. The pre-processed voice signal is then input into the ASR module for further processing.

[0037] Step 2: Automatic Speech Recognition:

[0038] The present invention uses automatic speech recognition technology to convert the preprocessed speech signal into text data. In the ASR module, the Paraformer-large non-autoregressive end-to-end speech recognition framework is used. The framework consists of an encoder, a predictor, a sampler, a decoder and a loss function, wherein the predictor and the sampler are the core components:

[0039] First: Predictor: It consists of a two-layer feedforward neural network (FFN), which is responsible for extracting the acoustic feature vector corresponding to the target text and accurately predicting the length of the target text.

[0040] Second: Sampler: It has no learnable parameters and converts the acoustic feature vector into a feature vector containing semantic information through sampling, which is further combined with the target text feature vector to enhance the bidirectional decoder's ability to model the context.

[0041] Step 3: Extract key information

[0042] First, the long dialogue text generated by the ASR module is deleting irrelevant special characters, punctuation marks, etc. in the long text, leaving only useful text content. After the text preprocessing is completed, the long text is split according to the preset segmentation criteria to obtain multiple short text segments T1,…,T i-1 ,T i The short text segments after splitting will be used as the basic processing unit in the subsequent key information extraction and generation process. By splitting the long text into multiple short text segments, it is ensured that the subsequent input of the text into the large language model will not be restricted due to the length being too long, and the input requirements are met. The specific segmentation criteria are as follows:

[0043] 3.1: The length of the short text segment is limited by the difference between the input length and the length of Prompt 1, that is, the length of the short text segment is not greater than the difference between the input length and the length of Prompt 1;

[0044] 3.2: When segmenting, divide by conversation rounds to ensure that one short text contains a complete conversation between the first user and the second user;

[0045] 3.3: If the current paragraph length does not reach the short text limit, continue to add the next round of dialogue content;

[0046] If the limit is exceeded, the current length is retained.

[0047] Then, the short text fragment and Prompt 1 are input into the large language model LLM in turn to generate key information:

[0048] T1→R 1,1 ,…,T i-1 →R 1,i-1 ,T i →R 1,i

[0049] The extraction results generated by LLM include to-do items, keywords, and text summaries. Next, take the result R of the first stage 1,1 ,…,R 1,m , and input LLM again with the second prompt word Prompt 2 to generate the second stage key information extraction result R 2,1 By analogy, R 1,n ,…,R 1,i After splicing, input LLM with Prompt 2 to get the result R of the second stage 2,j .

[0050] R 1,1 ,…,R 1,m →R 2,1 ;…;R 1,n ,…,R 1,i →R 2,j

[0051] It should be noted that the concatenation length is also limited by the difference between the input token length set by the system and the token length of Prompt2. n,i ) will be used as the input for the next stage, combined with Prompt2 and input into LLM again for processing. This process continues to iterate until the total length of the input text and Prompt2 can be input into LLM at one time, generating the final key information extraction result R.

[0052] The present invention is not limited to a specific LLM, and is applicable to open source models such as Qwen 2.5, LLaMA 3.1, GLM-4, and closed source models such as ChatGPT, ChatGLM, and Wenxinyiyan. From the above detailed description, it can be seen that since the iterative information extraction process is divided into input types of short text fragments and extracted fragment key information, Prompt1 and Prompt2 are designed separately. Specific examples of Prompt 1 and Prompt 2 are as follows Figure 2 and Figure 3 shown.

[0053] The above are only embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for extracting key information of long conversations in a 5G new call environment, characterized in that: The method comprises the following steps: Step 1: Voice signal collection and preprocessing: The system collects the voice signal between the first user and the second user through the 5G call device; Step 2: Automatic speech recognition: Automatic speech recognition technology converts the preprocessed speech signal into text data; Step 3: Extracting key information, the key information extraction includes the following sub-steps: Sub-step 31: splitting the long text according to a preset segmentation standard to obtain a plurality of short text segments after the segmentation; Sub-step 32: Input the short text segment and the first prompt word Prompt 1 into the large language model LLM to generate key information: the extraction results generated by the large language model LLM include to-do items, keywords and text summaries; Sub-step 33: Extract results by iteration: Take partial results of the first stage and concatenate them with the second prompt word Prompt2, input them into the large language model LLM again, and generate the key information extraction results of the second stage. 1,1 ,…,R 1,m After splicing, input LLM with Prompt 2 to generate the key information extraction result R of the second stage 2,1 By analogy, R 1,n ,…,R 1,i After splicing, input LLM with Prompt 2 to get the result R of the second stage 2,j . R 1,1 ,…,R 1,m →R 2,1 ;…;R 1,n ,…,R 1,i →R 2,j In each iteration, the output of the previous stage (R n,i ) will be used as the input for the next stage, concatenated with the second prompt word Prompt2, and then input into LLM again for processing; This process continues to iterate until the total length of the input text and the second prompt word Prompt2 can be input into LLM at one time, generating the final key information extraction result R; Step 4: Output result analysis.

2. According to claim 1, a method for extracting key information of a long conversation in a 5G new call environment is characterized in that: The length of the short text segment is limited by the difference between the input length and the length of the first prompt word Prompt 1.

3. According to claim 1, a method for extracting key information of long conversations in a 5G new call environment is characterized in that: The splicing length is limited by the difference between the input length set by the system and the length of Prompt2.

4. According to claim 1, a method for extracting key information of a long conversation in a 5G new call environment is characterized in that: The large language model LLM includes an open source model and / or a closed source model.

Images